|
|
 |
| Growth modeling with within-subjects ... |
 
|
| Message/Author |
|
|
| Ng Ee Lynn posted on Friday, February 26, 2016 - 12:21 am
|
|
|
Dear Drs. Muthen,
I would like to seek your advice on conducting growth modeling analysis involving within-subjects variables.
My study examines changes in cortisol levels in response to situational stress, and how change-in-cortisol predicts cognitive task performance. Participants completed a cognitive task under high and low situational stress (SS) conditions (i.e., situational stress was a within-subjects variable), and provided salivary cortisol at 4 time points at each level of SS. I would like to model the change/growth in cortisol separately for high and low SS, and to examine whether there are differences in the growth coefficients across different SS conditions. However, I'm not sure how to go about it. Should I: (a) specify separate growth models for each SS condition, or (b) specify one growth model and include SS as a covariate in the model?
I would appreciate any assistance or guidance, please. Thank you! |
|
|
|
|
You can do both
Multiple-group analysis for (a)
or (b).
MG gives a bit more modeling flexibility, but (b) is fine. |
|
|
| Ng Ee Lynn posted on Friday, March 04, 2016 - 12:27 am
|
|
|
Dear Dr. Muthen,
Thanks for your reply. I read up on multiple-group analysis and found that one of its key assumptions is independence of groups. Given that my "grouping variable" is situational stress (SS) conditions (high vs. low), would I be correct to say that this violates the independence assumption? If so, I suppose I should pursue option (b) instead?
For option (b), I'd given it some thought but am not completely sure how to implement the model in MPlus. Would you be able to share an example please?
Another option that I'd thought of is to first specify separate growth models for high and low SS. Then, to determine whether the relation between cortisol change and task performance (i.e., math accuracy) differs according to SS conditions, I would constrain the paths of interest (i.e., mathacc1 ON s1 and mathacc2 on s2) to be equal and run a chi-square difference test. An excerpt of the syntax is shown below:
MODEL:
!HIGH SITUATIONAL STRESS MODEL
i1 s1 | cort12@0 cort13@1 cort14*;
i1 WITH s1;
mathacc1 ON s1 (1);
!LOW SITUATIONAL STRESS MODEL
i2 s2 | cort22@0 cort23@1 cort24*;
i2 WITH s2;
mathacc2 ON s2 (1);
I would be grateful for your advice and insights please. |
|
|
|
|
| If that grouping means that different people are in different groups I don't see why it couldn't be done as multiple-group. |
|
|
| Ng Ee Lynn posted on Sunday, March 06, 2016 - 6:22 am
|
|
|
Dr. Muthen, my apologies for not describing my research design clearly in my previous posts. Please let me try again.
The same group of participants were exposed to both high and low situational stress (SS) conditions. What I mean is that in the syntax provided in my post on Friday, March 04, 2016 - 12:27 am, data for the HIGH SITUATIONAL STRESS MODEL and LOW SITUATIONAL STRESS MODEL come from the same participants. Hence, I posed the question about the viability of using multiple-group analysis to address my research questions.
I hope this helps to clarify the situation.
Thank you again for your time. |
|
|
|
|
| Groups in multiple group analysis must contain different individuals. |
|
|
| Ng Ee Lynn posted on Sunday, March 06, 2016 - 7:57 am
|
|
|
Dear Dr. Muthen,
Thanks for your quick reply. A follow-up question please:
1. In Dr. Bengt's reply on Friday, February 26, 2016 - 3:04 pm, he'd said that one option is to specify one growth model and include SS as a covariate in the model. However, I'm not sure how to specify the model in MPlus; my initial thoughts were to arrange my dataset in long-form (i.e., each participant has two rows of data, one row for each SS) and use the CLUSTER option to indicate that the dataset is clustered by participants. Am I on the right track? |
|
|
|
|
| Example 6.10 shows a growth model with covariates. |
|
|
| Back to top |
|
|
